TDPCA.m in TDPCA.

rar

function [eigvectors, eigvalues, meanData, newTrainData, newTestData] = TDPCA(trainData, testData, height, width, numvecs) %2DPCA Two Dimensional Principal component analysis % % % % % % % % % % % % % % % % % % % % % % Reference paper: J.Yang,D.Zhang,A.F.Frangi,and J.Yang.Two-dimensional pca:A new approach to a appearance-based face represenation and recognition. IEEE Trans.on PAMI,2004 Written by Zhonghua Shen ([email protected]), 2006.07 meanData: Mean of all the data. newTrainData: The data after projection (mean removed) newTestData: The data after projection (mean removed) eigvectors: Each column of this matrix is a eigenvector of the matrix defined in 2DPCA eigvalues: Eigenvalues of the convariance matrix trainData: Rows of vectors of training data points testData: Rows of vectors of testing data points height: height of the image matrix width: width of the image matrix numvecs: the needed number of eigenvectors Usage: [eigvectors, eigvalues, meanData, newTrainData, newTestData] =

TDPCA(trainData, testData, height, width, numvecs)

convariance

% Check arguments if nargin ~= 5 error('usage: [eigvectors, eigvalues, meanData, newTrainData, newTestData] = TDPCA(trainData, testData, height, width, numvecs)'); end; [nSam,nFea] = size(trainData); fprintf(1,'Computing average matrix...\n'); meanDataVector = mean(trainData); meanData = reshape(meanDataVector,height,width); fprintf(1,'Calculating matrix differences from avg and 2DPCA covariance matrix L...\n');

TDPCA.m in TDPCA.rar

L = zeros(width,width); ddata = zeros(nSam,nFea); for i = 1:nSam ddata(i,:) = trainData(i,:)-meanDataVector; dummyMat = reshape(ddata(i,:),height,width); L = L + dummyMat'*dummyMat; end; L = L/nSam; L = (L + L')/2;

fprintf(1,'Calculating eigenvectors of L...\n'); [eigvectors,eigvalues] = eig(L); fprintf(1,'Sorting eigenvectors according to eigenvalues...\n'); [eigvectors,eigvalues] = sortem(eigvectors,eigvalues); eigvalues = diag(eigvalues); fprintf(1,'Normalize Vectors to unit length, kill vectors corr. to tiny evalues...\n'); num_good = 0; for i = 1:size(eigvectors,2) eigvectors(:,i) = eigvectors(:,i)/norm(eigvectors(:,i)); if eigvalues(i) < 0.00001 % Set the vector to the 0 vector; set the value to 0. eigvalues(i) = 0; eigvectors(:,i) = zeros(size(eigvectors,1),1); else num_good = num_good + 1; end; end; if (numvecs > num_good) fprintf(1,'Warning: numvecs is %d; only %d exist.\n',numvecs,num_good); numvecs = num_good; end; eigvectors = eigvectors(:,1:numvecs); if nargout == 5 fprintf(1,'Feature extraction and calculating new training and testing data...\n'); newTrainData = zeros(nSam,height*numvecs); for i = 1:nSam

TDPCA.m in TDPCA.rar

dummyMat = reshape(ddata(i,:),height,width); newTrainData(i,:) = reshape(dummyMat*eigvectors,1,height*numvecs); end %testData nt = size(testData,1); newTestData = zeros(nt,height*numvecs); tdata = zeros(size(testData)); for i = 1:nt tdata(i,:) = testData(i,:)-meanDataVector; dummyMat = reshape(tdata(i,:),height,width); newTestData(i,:) = reshape(dummyMat*eigvectors,1,height*numvecs); end; end;

Image Feature Extraction based on PCA..

function [XP,W,eigv,sumeigv]=pca(X,m) %input: % X [d * n] after truncation hyperspectral data % M [1 * 1] After the dimensionality reduction of data dimension %output: % W [d * m] transformation matrix % Eigv [1 * m] input data covariance matrix eigenvalues (from the large inverted small order) % XP [m * n] transformed principal component weight %load X.mat; K=cov(X'); [U, D] = eig (K); [Eigv, index] = sort (diag (D), 'descend'); U = U (:, index); % compute eigenvectors of cov(X) %m=6; if m> length (eigv) disp('Desired data dimension is too big.'); V = U; else V = []; for i=1:m eigv if (i)> 0 V = [V, U (:, i) / sqrt (eigv (i))]; end end end eigv eigv = (1: m); sumeigv = sum (eigv (m +1: end)); W = V / norm (V); XP=W'*X; return; ...

TDPCA.m in TDPCA.rar

pca source for feature extraction using Matlab language, pca that Principal Comp

% A revised PCA for face recognition in Matlab code, in fact, I feel that feature extraction and classification using the coordinates of the feature space % Calc xmean, sigma and its eigen decomposition allsamples =[];% all training images for i = 1:40 for j = 1:5 a = imread (strcat ('D: \ rawdata \ ORL \ s', num2str (i), '\', num2str (j), '. pgm')); % Imshow (a); b = a (1:112 * 92);% b is the row vector 1 N, in which N = 10304, extraction order is then over, that is from top to bottom, left to right b = double (b); allsamples = [allsamples; b];% allsamples is a M * N matrix, allsamples representative of each row of data in a picture, in which M = 200 end end samplemean = mean (allsamples);% average picture, 1 N for i = 1:200 xmean (i,:) = allsamples (i,:)-samplemean;% xmean is an M N matrix, xmean save each line of data is "data for each image - an average picture." end; sigma = xmean * xmean ';% M * M-order matrix [Vd] = eig (sigma); d1 = diag (d); [D2 index] = sort (d1);% in ascending order cols = size (v, 2);% eigenvector matrix of columns for i = 1: cols vsort (:, i) = v (:, index (cols-i +1));% vsort is a M * col (Note: col generally equal to M)-order matrix, is stored in descending order of feature vectors, form a feature vector for each column dsort (i) = d1 (index (cols-i +1));% dsort is stored in descending order of eigenvalues, is a one-dimensional row vector end% complete in descending order % Of 90% of the energy options dsum = sum (dsort); dsum_extract = 0; p = 0; while (dsum_extract / dsum <0.9) p = p + 1; dsum_extract = sum (dsort (1: p)); end i = 1; % (Training phase) calculated the formation of the face coordinate system feature while (i <= p & & dsort (i)> 0) base (:, i) = dsort (i) ^ (-1 / 2) * xmean '* vsort (:, i);% base is the N p order matrix, divided by dsort (i) ^ (1 / 2) is a face image of standardization, see "PCA-based face recognition algorithms" p31 i = i + 1;

TDPCA.m in TDPCA.rar

end % Add by wolfsky is the following two lines of code, the training samples on the projection of the coordinate system, to get a M * p-order matrix allcoor allcoor = allsamples * base; accu = 0; % Test for i = 1:40 for j = 6:10% Read 40 x 5 test images a = imread (strcat ('D: \ rawdata \ ORL \ s', num2str (i), '\', num2str (j), '. pgm')); b = a (1:10304); b = double (b); tcoor = b * base;% calculate the coordinates, is 1 p order matrix for k = 1:200 mdist (k) = norm (tcoor-allcoor (k ,:)); end; % Third-order neighbors [Dist, index2] = sort (mdist); class1 = floor (index2 (1) / 5) +1; class2 = floor (index2 (2) / 5) +1; class3 = floor (index2 (3) / 5) +1; if class1 ~ = class2 & & class2 ~ = class3 class = class1; elseif class1 == class2 class = class1; elseif class2 == class3 class = class2; end; if class == i accu = accu +1; end; end; end; accuracy = accu/200% recognition rate of output